Current research on hospitalized patients has focused on patients in intensive care units (ICUs), which have been early adopters of electronic medical records (EMRs). Research on general medical-surgical ward patients has been limited due to the high cost of manual abstraction of physiologic data, particularly vital signs. Given the paucity of such data, clinicians lack quantitative tools to gauge the process of hospital care at different points in time, let alone the level of risk of deterioration a given patient may have at different points in the course of a hospital stay. This project employs the inpatient EMR to provide such tools. It focuses on a specific patient population that has extremely high morbidity and mortality: hospitalized adults with community-acquired pneumonia (CAP) who experience an unplanned transfer to a higher level of care (e.g., from a general medical surgical ward to the ICU). Approximately 70% of these transfers occur in the first 72 hours in the hospital, and death rates among these patients range from 10 to 40%, with severity-adjusted observed to expected mortality ratios as high as 16. Our long term goal is to harness the power of the inpatient EMR for quality monitoring, quality improvement, and the identification of effective practices and interventions designed to prevent in-hospital deterioration. To achieve this goal, we have two specific aims: (1) Using a case-cohort methodology, we will develop models, suitable for embedding in the EMR, to predict the occurrence of critical illness within 72 hours of hospital admission among CAP patients who were not initially admitted to the ICU. Using comprehensive inpatient and outpatient EMR data from 20 Northern California Kaiser Permanente hospitals, we will identify a cohort of approximately 13,700 hospitalized patients meeting the following criteria: age =18 years, admission diagnosis of CAP;and not admitted only for palliative or comfort care. Critical illness is defined as (a) shock, (b) respiratory failure requiring assisted ventilation, and/or (c) cardiac arrest. We will develop predictive models using approximately 8,800 patient hospitalization records (of which we estimate 485, or 5.5%, will develop critical illness within 72 hours) and validate them on approximately 4,900 patient records (with 270 developing critical illness within 72 hours). (2) Using the results of Specific Aim 1, we will generate time-delimited "snapshots" of the characteristics of CAP patients who did and who did not develop critical illness. The "snapshots" will characterize CAP patients on admission and at 12, 24, and 48 hours into their hospital stay with respect to (a) their demographic, clinical, and physiologic characteristics (including vital signs, laboratory test results, and severity of illness scores);(b) key processes of care, such as whether and when specific tests (e.g., chest roentgenograms, pulse oximetry, lactates) and interventions (e.g., provision of supplemental oxygen, treatment with systemic antibiotics, intravenous fluid boluses) were performed;and (c) their hospital outcomes, including deterioration, death, length of stay (LOS), and discharge disposition for survivors.